We introduce SCFlow2, a plug-and-play refinement framework for 6D object pose estimation. Most recent 6D object pose methods rely on refinement to get accurate results. However, most existing refinement methods either suffer from noises in establishing correspondences, or rely on retraining for novel objects. SCFlow2 is based on the SCFlow model designed for refinement with shape constraint, but formulates the additional depth as a regularization in the iteration via 3D scene flow for RGBD frames. The key design of SCFlow2 is an introduction of geometry constraints into the training of recurrent matching network, by combining the rigid-motion embeddings in 3D scene flow and 3D shape prior of the target. We train SCFlow2 on a combination of dataset Objaverse, GSO and ShapeNet, and evaluate on BOP datasets with novel objects. After using our method as a post-processing, most state-of-the-art methods produce significantly better results, without any retraining or fine-tuning. The source code is available at https://scflow2.github.io.

Medical image segmentation is a key technology in computer-aided diagnosis, but existing methods face an imbalance between efficiency, accuracy, and generalization capabilities. Despite significant advances in CNNs, Transformers and SSMs, these approaches still encounter fundamental challenges in model design, particularly in parameter efficiency. In this paper, we propose the Efficient Lightweight Network with the spaceLink block. Specifically, we design the Channel-Attention Feature Refinement Module (CAFRM) with Space-Link Block (SLB), dramatically reducing parameters compared to other methods. The CAFRM method emphasizes important features, while the SLB method reduces computational cost with depthwise separable convolutions and shares spatial parameters. Experimental results demonstrate that our approach requires around 15 K parameters while maintaining competitive performance, significantly reducing computational costs compared to existing methods and making it particularly suitable for resource-constrained medical environments.

Conventional multi-kernel clustering methods typically focus on selecting diverse kernel functions in the original feature space while overlooking the structural information embedded in different frequency components. This limits their ability to handle complex data distributions, especially with multi-scale patterns and noise. To address this, we propose a novel clustering paradigm that leverages Beta wavelet transforms to construct frequency-aware kernel representations. By systematically decomposing single-view data into multiple frequency components, our method enriches the feature space with hierarchical spectral information. Beta wavelet filtering generates transformed representations that emphasize distinct frequency characteristics, capturing both local and global structures. A multi-kernel fusion strategy integrates these representations, preserving scale-specific similarities while maintaining cross-frequency coherence. Experiments on benchmark datasets demonstrate notable improvements in clustering accuracy, validating the benefit of incorporating frequency-domain insights into multi-kernel clustering. This work introduces a new perspective on unsupervised learning, highlighting spectral decomposition as a key step for robust and interpretable clustering. Code is available at https://github.com/Lwhite111/BWMSKC.

Object detectors for remote sensing imagery (RSI) are plagued by data-driven training approaches, underperforming when confronted with inadequate object information. Thus, semi-supervised object detection (SSOD) is introduced to alleviate high data dependence. However, it’s unfeasible to directly apply existing SSOD methods, considering the classification and regression uncertainties resulting from unique characteristics of RSI: 1) inter-class uncertainty: objects in remote sensing scenes are usually densely arranged, leading to severe feature fusion between different categories, 2) intra-class uncertainty: for same-category objects, the complicated contexts induced by various scenarios increase intra-class feature diversity, posing a challenge to clustering and 3) regression uncertainty: the inherent scale variation in RSI gives rise to inconsistent convergence rate, hindering the regression progress from effective converging. Therefore, we propose a Teacher-Student Model(TSM)-based SSOD method aiming at remote sensing scenes, termed Uncertainty-Aware Teacher (UAT), to quantify and promote the detector’s confidence, which is composed of Entropy-based Uncertain-pair Softening (EUS) policy, Multi-Gaussian Sample Fitting (MSF) module and Scale-adaptive Adjunct Loss (SAL). Specifically, EUS identifies the uncertain category pairs with class-wise information entropy, MSF remodels the intra-class distribution and sets the optimal threshold dynamically for pseudo-labels, and SAL adaptively re-scales bboxes based on the original size to accelerate the regression convergence. We have proven the efficiency of our method through extensive experiments on two public remote sensing datasets, DOTA and DIOR.